Pedestrian Detection in Haze Environments Using Dark Channel Prior and Histogram of Oriented Gradient

Abstract

Pedestrian detection is a critical requirement for industry computer vision systems and has great research value. Many techniques have been proposed, and one of the most effective methods is based on the Histogram of Oriented Gradient (HOG) descriptor and the Support Vector Machine (SVM) classifier. While this method implicitly assumes that the input images are taken in haze-free environments. In this paper, we propose a new method that is capable of handling pedestrian detection in haze environments. Firstly, a haze version of 'INRIA' data set is synthesized based on the natural light transport model. Secondly, robust HOG descriptors are calculated by a fusion model which relies on the dark channel prior haze removal method. At last, a linear SVM classifier is trained for pedestrian detection. The proposed method outperforms the traditional HOG pedestrian detection method on the haze version of 'INRIA' data set.